Enterprise Machine

Learning Guide
The executive’s playbook for business
success with machine learning
INTRODUCTION TO MACHINE LEARNING

Machine learning is quickly becoming a fundamental

building block of business operations, providing 63%
real impact such as improved processes, increased of respondents report revenue
increases in the business units where
efficiency, and accelerated innovation. Advances their companies use AI 2
in machine learning technology have combined
with high-performance compute options and an
abundance of data to create a perfect storm to
transform organizations of all sizes.

For enterprise leaders, the need to embrace machine learning capabilities is

immediate. Gartner forecasts that AI-derived business values are projected
to reach $3.9 trillion in 2022.1 As enterprises accrue data from numerous
sources—ranging from customer feedback and sales logs to internal
processes and financial forecasting—there is an opportunity to operationalize
this often “dark data” into insights that can create new revenue streams.

1
Petty, C., & van der Meulen, R. (Eds.). (2018, April).
2
https://www.mckinsey.com/featured-insights/artificial-intelligence/global-ai-survey-ai-proves-its-
worth-but-few-scale-impact

2
2
GOALS

Impacting business goals

Companies are using machine learning across the organization to address
business priorities. Some machine learning initiatives tackle incremental gains
to automate processes to create efficiencies. Others are transformational
initiatives aimed at innovation and competitive differentiation. While there are
numerous applications of machine learning, enterprises already realize great
value from use cases that provide new experiences for their customers and
drive business growth.

Goals for machine learning innovation

1 Enable product and service innovation

2 Drive research and discovery

3 Enhance the customer experience

4 Modernize customer service

5 Increase efficiency and productivity

6 Improve security and compliance

7 Optimize operations

8 Enable agile decision making

3
GOALS

1 Enable product and 2 Drive research and discovery

service innovation In industries such as automotive, manufacturing,
Enterprises are already using machine learning and energy and utilities, machine learning helps
pervasively to enable product and service uncover new formulas, new materials, and new
innovation. They inform their product roadmap ways of thinking about age-old problems. It is
through intelligence extracted from customer especially relevant in industries such as healthcare
feedback; drive the product development and life sciences, where machine learning models
lifecycle, including DevOps and quality assurance can make predictions that drive research toward
through automation and intelligence; and infuse breakthroughs that lead to new and improved
machine learning capabilities directly into new patient treatments.
products that benefit the end user.

3 Enhance the 4 Modernize customer service

customer experience Critical to every business is ensuring that
Across industries including financial services, various customer touch points—both internal
healthcare, retail, and media and entertainment, and external—provide timely, accurate, and
targeted recommendations have become meaningful customer service. Machine learning
important for personalization that delights and technology such as conversational agents bridge
engages customers. Epsilon research indicates the gap, modernizing the contact center to
80 percent of consumers are more likely to make provide quick responses to customer calls and
a purchase when brands offer personalized chat requests. In addition, text-to-speech and
experiences.3 Using voice-of-customer natural language processing capabilities can be
analytics to understand customer behavior and applied to the streams of incoming customer
preferences, as well as recommendations engines data such as texts, voice messages, and customer
to predict best matches, enterprises can push service logs to better understand customer needs
customized offers to location-based mobile and sentiment to help improve the quality
services that offer real-time convenience of customer service.
and choice.

Epsilon. “The Power of Me: The Impact of Personalization on Marketing

3

Performance.” Epsilon, us.epsilon.com/power-of-me.

4
9 GOALS

5 Increase efficiency 6 Improve security

and productivity and compliance
Across business operations, the pervasive Use of machine learning across security, risk, and
use of machine learning to automate and compliance use cases is a fast-growing trend,
streamline processes has resulted in increased especially in the financial services sector. Fraud-
efficiency and reduced costs. For example, detection models help to keep consumer data
using automated media tagging and large- safe and prevent malicious attacks against bank
scale document recognition and analysis, accounts or even mobile end points. On the other
machine learning can eliminate countless hours end of the spectrum, machine learning automates
of labor. Machine learning can also analyze mundane tasks such as financial-document
manufacturing incidents in real time, providing analysis, reducing manual effort and allowing the
early warning of potential problems. workforce to focus on higher-value tasks.

7 Optimize operations 8 Enable agile decision making

From point of sale to freight delivery Machine learning-based predictions augment
management to forecasting demand, machine decision making across all functions of the
learning plays a pivotal role across the supply organization and across use cases. These predictions
chain. In demand planning, for example, machine produce continuous actionable insights that help
learning algorithms can look at historical data as leaders and teams tackle operational and business
well as demand and other relevant data—such challenges. This can range from forecasting that
as product schedules, weather patterns, and impacts pricing optimization to predictions for
competitor pricing—to determine when it’s time autonomous vehicles to make informed choices.
to restock or end of life a product. Sophisticated In healthcare, behind the scenes, machine learning
machine learning models can predict weekly, is being used to analyze everything from x-rays to
daily, and even down-to-the-minute forecasts patientsʼ history data, helping clinicians make better
to help companies streamline supply chain and faster decisions.
management, save costs, and increase efficiency.

5
CHALLENGES

Challenges of adoption
Machine learning is frequently the catalyst that turns business data into and iterating over and over again. There are decisions to be made about
accurate predictions and actionable information, but as with many emerging infrastructure—selecting the right compute for training and inference,
technologies, there are challenges to adoption, including data, complexity, considerations for cloud, on-premises, and edge deployments.
cost, and lack of skills.
Cost inhibitors
Data ambiguity
Machine learning training and inference can be expensive, especially since
Enterprises can struggle with various issues related to data. First and models require iterations to improve the accuracy of predictions. Because
foremost, many are unaware of all their possible data sources that might hold embarking on machine learning initiatives is new to many companies, they
hidden insights. Even when they’ve identified data, there’s a lack of labeled also don’t have the experiences or skills in-house and often have to rely on
data ready for machine learning. Furthermore, even labeled data can prove costly external resources to kick-start projects.
to be an issue where integrity is in question since data can often have hidden
biases based on human labelers. Finally, enterprises often struggle with Lack of skills
ensuring the right data management and governance policies are in place to
Even when companies embrace new technologies like machine learning to
allow the right people and processes to securely access, store, and manage
drive business transformation, having the right skills is often a roadblocker
the data.
for getting started. Machine learning initiatives require machine learning
expertise to build and train machine learning models—this includes the skills
Complexity
of machine learning developers, data scientists, and researchers to build
The machine learning workflow can be time-consuming and iterative, which algorithms and train models. These skills are not in great supply and are
leaves many organizations and developers thinking machine learning is often unavailable in-house, which leaves businesses struggling to train or
complex and difficult to use. There are many steps involved, from prepping supplement skills with consultants and partners.
data and choosing algorithms to building, training, and deploying models…

6
KEY CONSIDERATIONS

Getting started with machine learning Key considerations

Advance your data strategy

Machine learning presents new opportunities to
Understand business objectives
realize foundational gains such as efficiency and
Leverage the cloud
cost saving, as well as higher-value gains such as
Enable your organization
product innovation and spurring discovery and
research. But how do organizations get started?
For many, machine learning adoption begins by
considering all data sources and existing data
strategies. They identify workflows and business
processes that suffer from low efficiency. They
consider all their data sources and existing data
strategy. They determine the best cloud-based
infrastructure and tools to scale machine learning.
And last, they ensure that the right skills are on
board for the machine learning initiative to be
successful.

7
KEY CONSIDERATIONS

Data is gold for leaders

who are looking to disrupt
their industries

Advance your data strategy Understand business objectives

Data is gold for leaders who are looking to disrupt their industries with Understanding the business benefits of machine learning adoption—in
machine learning. But many organizations don’t have machine learning- particular, the specific benefits relevant to your organization—is critical for
ready data. Recognizing the importance of data and developing a plan to enterprise success with machine learning. Once objectives are identified,
collect and use that data is critical for successful machine learning adoption, it’s important for business and technical leaders to understand and
even at the proofs of concept (PoC) stage. champion roles.

All sources of data need to be uncovered, from structured data like billing Select a targeted use case
and CRM to unstructured data like social media feeds, images, and forms. When choosing a pilot, consider use cases where machine learning can
Then that data needs to be evaluated for quality and usefulness. Finally, have the most impact and those from which you can learn to scale
data needs to be cleaned and accurately labeled for machine learning enterprise deployment. Focus on how you can deliver a better experience
models to transform it into valuable insights. for your customers and identify the business and operational outcomes
desired. Then, establish one or two high-value PoCs that can really make
a difference to your organization and quickly demonstrate results. For the
PoC to succeed, it’s critical to have the right resources in place, including
infrastructure, data, and capabilities.

8
KEY CONSIDERATIONS

Understand the impact

From the outset, consider the operational effect of new machine learning
solutions. Machine learning can have a transformative impact, so it’s
important to plan in advance for what you want to achieve and measure.
This can also help you determine how to measure success. Rather than
aiming for a target financial ROI within a given time frame, you’ll find more
success by measuring the impact of machine learning initiatives in terms of
business agility, competitive advantage, and risk tolerance.

Iterate and learn

Once you’ve proven the potential of machine learning, the next step is to
move from pilot to production, which may include integrating the machine
learning capability into a larger IT system. This move typically takes longer
than the pilot process and can vary depending on the complexity of the
overall system and how large-scale the production deployment will be.

Leverage the cloud

Successful machine learning initiatives need more than just the right tools.
A comprehensive platform brings together data store, security, and analytics
services, as well as compute resources for training and deployment. Turning
to the cloud for these services brings a wide range of benefits, including
speed, scalability, flexibility, resilience, security, and reduced cost.

Additionally, the cloud offers the widest range of high-performance CPU

and GPU processor types, which are essential for large-scale training and
for deployment in a production environment. Using cloud-based data lakes
and storage also ensures that you can easily access and manage data so that Identify desired outcomes and
machine learning initiatives are seamless, repeatable, and scalable.
deliver better experiences for
your customers

9
KEY CONSIDERATIONS

Enable your organization Enlist the right stakeholders

Along with the right use cases, having the right skills to build machine learning and skills to get pilots off
applications and systems, as well as the right process and operating model, the ground and scale
is essential to getting pilots off the ground and scaling enterprise machine learning.

Assemble the team

Assemble a team of machine learning developers and data scientists essential for a
successful PoC, and train teams for future deployments. It’s also important to involve
subject matter experts who understand your business vernacular, especially for industry
domains, to help you get to the ground truth with your data. Consider appointing a
Chief Data Officer (CDO) to lead the charge on data strategy and governance, bring
together interdisciplinary teams, and streamline data processes.

Create the process

Machine learning may not bring the expected value if the results are not integrated
with other areas of the organization. Operationalizing machine learning models
is hard—as many as half of PoCs don’t get deployed into production. Therefore,
executive sponsorship to change business processes and alignment with application
development is key. Successful teams create processes to align machine learning
experts, data scientists, and developers with key business stakeholders. A well-defined
process also helps ensure the final output is well integrated into business processes.

Build the culture

To help realize its potential, there needs to be cultural acceptance that machine
learning is an important part of business and operations. Some initiatives may
require information from across these domains, so it’s important to understand all
the stakeholders who need to be involved, and bring together stakeholders who can
champion adoption.

10
CUSTOMER FOCUS

Disney uses ML to sort through a

universe of content
With nearly a century of content on its hands, a growing percentage
of it digital, Disney needed to organize its vast library more carefully
than ever.

The answer was metadata: information about stories, scenes, and

characters. For example, Bambi would have metadata tags that
identify not only characters like Thumper but also the type of
animal, relationships between animals, the character archetypes
each animal portrays, and things like the nature scenes, music,
sentiment, and tone of the story. Appropriately tagging all of this
content with the right metadata is challenging.

Disney is building machine learning and deep learning tools with

the help of AWS to automatically tag this content with descriptive
metadata to make the archiving process more efficient. As a result,
writers and animators can quickly search for everyone from Mickey
Mouse to Modern Family’s Phil Dunphy.

Learn how Disney uses machine learning to keep its

archive accessible to writers and illustrators seeking
inspiration »

11
CUSTOMER FOCUS

Formula 1 accelerates cloud

transformation with ML
Formula 1 is a data-driven sport: Each race, 20 sensors on each car
generate 3 GB of data, and 1,500 data points are generated each
second. Using Amazon SageMaker, Formula 1’s data scientists are
training deep learning models with 65 years of historical race data to
extract critical performance statistics, make predictions, and give fans insight
into the split-second decisions and strategies adopted by race teams.

By streaming real-time race data to AWS using Amazon Kinesis,

Formula 1 can capture and process key performance data for each car
during every twist and turn of the Formula 1 circuits. Sharing these
insights through television broadcasts and digital platforms allows
Formula 1 to give fans access to the inner workings of their favorite
teams and drivers.

“By leveraging Amazon SageMaker and

“AWS outperforms all others in speed, scalability, reliability, global reach,
partner community, and breadth and depth of cloud services,” said Pete AWS’s machine learning services, we are
Samara, director of innovation and digital technology at Formula 1. “By able to deliver these powerful insights and
leveraging Amazon SageMaker and AWS’s machine learning services, we are
able to deliver these powerful insights and predictions to fans in real time.” 
predictions to fans in real time.” 

Pete Samara, Director of Innovation and Digital Technology, Formula 1

Learn more about Formula 1 and AWS
machine learning services»

12
CUSTOMER FOCUS

Domino’s Pizza Enterprises Limited

accelerates order and delivery with ML
Domino’s Pizza Enterprises Limited is an increasingly digital business,
with more than 70 percent of sales coming from online orders. To
enable faster pickup and delivery, the company launched an initiative
that aims to have a pizza ready for pickup within 3 minutes or safely
delivered within 10.

Domino’s turned to AWS to help create a predictive ordering solution,

building a data lake consisting of key order information by taking
advantage of Amazon S3 for data storage and AWS Glue for data
querying. It also uses Amazon SageMaker to build and train machine
learning models to predict the likelihood that an order will be placed.

Domino’s is delivering pizza faster, hotter, and fresher with AWS

“Customers are getting their pizza
machine learning capabilities. Predictive ordering gives Domino’s faster, hotter, and fresher because of the
a competitive advantage in the marketplace, compared to some
improvements we’ve put into place with
competitors who take 45 minutes to an hour to deliver a meal.
Project 3TEN. The predictive ordering
Learn how Domino’s leverages machine learning for solution we developed by using AWS is a
predictive ordering to deliver to hungry customers faster » big part of that.”

Michael Gillespie, Chief Digital and Technology Officer, Domino’s

Pizza Enterprises Limited

13
CUSTOMER FOCUS

GE Healthcare drives better

outcomes with ML
The healthcare sector is poised to benefit significantly from the
innovations of big data, machine learning, and artificial intelligence.
Leading the way is GE Healthcare, which has partnered with the
University of California, San Francisco to create a library of deep
learning algorithms centered around improving traditional x-ray
imaging technologies like ultrasounds and CT scans.

The goal is to allow physicians to treat patients more quickly, not

only cutting costs but also improving outcomes. Behind it is GE
Healthcare’s partnership with AWS, which provides the ability to
“We want to use the AWS platform to scale deploy machine learning solutions at scale through the Amazon
to as many algorithms as we can. The more SageMaker machine learning platform.

intelligence we can put into medical devices

Read how GE Healthcare and AWS have
and applications, the more we can increase
partnered to make goals a reality»
the quality. It’s going to improve access, it’s
going to improve efficiency and it’s going to
reduce costs all at the same time.”
Keith Bigelow, General Manager of Analytics, GE Healthcare

14
GET STARTED WITH AWS ML

AWS ML and AI Services

AWS has the broadest and deepest set of machine learning (ML) and AI services for your
business. On behalf of our tens of thousands of customers, we are focused on solving some of
the toughest challenges that hold back machine learning adoption.

200+
Choose from pre-trained machine learning services such as Amazon SageMaker to build and
scale machine learning, or build custom models with support for all the popular open-source
frameworks. AWS machine learning capabilities are built on a comprehensive cloud platform,
optimized for machine learning with high-performance compute, security, and analytics. ML features and services
in the last 12 months

ML Services
Amazon SageMaker provides every developer and data scientist with the ability to build, train,
and deploy machine learning models quickly, without needing data engineers or DevOps.
Amazon SageMaker is a fully managed service that covers the entire machine learning workflow
75%
to label and prepare your data, choose an algorithm, train the model, tune and optimize it for Lower ML
deployment, make predictions, and take action. Your models get to production faster with inference cost
much less effort and lower cost.

AI Services
AWS pre-trained AI Services provide ready-made intelligence for your applications and
10x
Better ML algorithm
workflows. AI Services easily integrate with your applications to address common use cases
performance
such as personalized recommendations, modernizing your contact center, improving safety
and security, and increasing customer engagement. Because we use the same deep learning

1-click
technology that powers Amazon.com and our Machine Learning Services, you get quality and
accuracy from continuously learning APIs. And best of all, AI Services on AWS don’t require
machine learning experience.
Model training
and deployment

15
GET STARTED WITH AWS ML

Amazon ML Solutions Lab

The Amazon ML Solutions Lab combines hands-on educational workshops
with advisory professional services to help you ‟work backward” from
business challenges, and then go step-by-step through the process of
developing machine learning-based solutions. You’ll be able to take what
you have learned through the process and use it elsewhere in your
organization to apply machine learning to business opportunities.

AWS Machine Learning Competency Partners

AWS Machine Learning Competency Partners offer a range of services and
solutions to help you create intelligent solutions for your business. From
concept development through execution, partner solutions offer tools and
expertise to help with the entire machine learning workflow.

Training and Education

Start training on machine learning on AWS with courses based on the
same material used to train Amazon’s developers through the combination
of foundational knowledge and real-world application. Developers, data
scientists, data platform engineers, and business decision-makers can use this
training to learn how to apply machine learning, AI, and deep learning to
their businesses.

Learn how to put machine learning

to work for your business »

16
© 2021, Amazon Web Services, Inc. or its affiliates. All rights reserved.